The Situation Room is brought to you by Trusted Agents​

Free Cancellation Was Built for Humans

The Missing Layer Is Agentic Resource Planning

If you gave a personal travel agent permission to save you money, what would it actually do?

Most people assume the polite version: it watches prices, waits patiently, and books once when the price is right.

Agents do what humans don’t.

As Jamie puts it, “operations have been designed for human scale transactions and decision making. And if the market offers refundable inventory, the rational thing for software to do is not patience. It’s optimisation.”

So if your agent is allowed to optimise for price and flexibility, and it has access to refundable rates, it behaves like a portfolio manager. It books early to secure a position. Then it keeps shopping. It books a better option when it appears. It cancels the weaker one at the last responsible moment. It repeats the cycle until the cancellation window closes.

Humans do a little of this already, but it stays bounded because humans have friction. They get tired. They get busy. They hesitate. They don’t want to feel like they’re gaming the system and they worry that they might forget to cancel.. They also can’t hold ten overlapping bookings in their head while comparing room class, location, cancellation terms, breakfast, loyalty benefits, and price trajectories across multiple channels.

Agents can.

And if there’s a loophole in the commercial design, they will find it. Not because they’re malicious, but because they’re doing exactly what we asked: minimise price, maximise flexibility, avoid regret.

Evidence it already exists today

This isn’t a future edge case. “Book and search” behaviour is already well understood in hospitality: customers book to secure availability, keep shopping, then cancel and rebook when they find something better. Zvi Schwartz describes how cancellations and no-shows undermine the two pillars of revenue management, forecasting and control, and why this behaviour creates noise in the booking curve that makes pricing decisions worse.

The Cancel, rebook, save paper puts hard numbers behind it. The authors analyse 2,223,024 reservations across 628 Portuguese properties (2022–2024) and estimate rebookings accounted for 0.94% of reservations, producing €1,241,281.58 in gross revenue displacement. They also note the operational problem: rebookers can look like normal guests early in the journey, which makes the behaviour harder to detect and manage in real time.

So the point is simple: agents didn’t invent this behaviour but they will industrialise it.

Why this should keep hospitality operators awake

Once refundable inventory becomes machine-arbitrageable, your demand signals stop being trustworthy. The booking curve starts to reflect “optioned intent” rather than committed intent, and that changes what your revenue system thinks it knows.

You see it first in forecasting. Early pickup looks strong, so you tighten inventory, push prices, or shift channel allocation, only to watch a chunk of that demand evaporate close to arrival when the agent collapses its portfolio. What comes back to you late is often the hardest inventory to resell at the best rate, and it shows up at the moment your levers are weakest.

Then comes pricing feedback. If many agents are running the same play, your dynamic pricing starts reacting to phantom demand, and your own price cuts become a trigger for automated rebooking. That creates a loop where “doing the right thing” for short-term conversion can quietly increase long-term revenue leakage.

And it hits operations, not just revenue. More changes. More cancellations. More refunds. More customer service contacts, especially from guests who didn’t personally make the decision and therefore have less tolerance when anything deviates. In an agent-mediated world, errors and exceptions feel less like inconvenience and more like betrayal, because the promise was made by software on the customer’s behalf.

This is why agentic travel isn’t just a new interface problem. It’s a new contract problem. You’ll need to decide what a reservation really represents: a free option, or a commitment with consequences. That means deliberately redesigning refundable inventory, cancellation windows, and the signals an agent must provide when it books, changes, or cancels.

The View from 3 Seats

A quick sanity check from three vantage points, because this issue looks very different depending on where you sit:

Hotel / operator: Your revenue system was trained on human booking curves. If bookings become low-cost options held by software, pickup stops being a reliable signal, and late cancellations become an operational event, not a rounding error.

Intermediary (OTA/GDS/metasearch): The player with the widest cross-market visibility can see true intent emerging from all the “options,” and can steer demand, pricing pressure, and merchandising in ways individual properties cannot.

Traveller / personal agent: The agent is doing exactly what it was asked to do: minimise price, maximise flexibility, reduce regret. If the supplier-side rules make that easy, the agent will keep exploiting it, quietly and at scale.

The missing layer: Agentic Resource Planning

Most of the agentic travel conversation sits at the top of the stack. Better discovery. Better planning. Better recommendations. That’s the photogenic part.

The harder part is what happens after the agent clicks “book.” In an agent world, the work shifts from helping humans navigate screens to running a system that can handle machine-speed change: bookings that behave like options, constant optimisation, and high-volume servicing events that land all at once near the cancellation deadline.

This is the layer we think is missing from most roadmaps: Agentic Resource Planning. Not a new chatbot. The operational control plane that makes agent demand governable and fulfilment predictable. In travel, that means treating inventory, pricing, servicing, and exceptions as agent-readable systems, with clear commitment signals and limits.

You can see the same pattern in retail logistics. Anurag Singh puts it crisply: AI agents can find, compare, and buy in seconds, but they can’t deliver the package. The back end decides whether the promises made at the front end are real.

Translate that to hospitality and travel ops. Your equivalent of the 3PL (3rd Party Logistics) is the unglamorous machinery that makes promises true: CRS and RMS logic, channel managers, payment and fraud controls, customer service, refund workflows, and the systems that manage disruption. If those are not designed for agent behaviour, you end up with two outcomes: the agent routes demand elsewhere because you cannot confirm and service reliably, or you accept demand and then drown in changes and exceptions.

Agentic Resource Planning is how you avoid both. It is the set of design choices that make “agent time” survivable: commitment-weighted forecasting, tiered cancellation design, limits on repeated cancel and rebook patterns, machine-readable servicing surfaces (change, refund, dispute), and evidence trails that let you explain what happened when something goes wrong.

Ask the Hard Question

You can treat this as edge-case behaviour and hope it stays small. Or you can assume it scales, because the incentives are obvious and the tooling is getting cheaper every month.

The practical implication is that you may need to redesign how bookings work when the buyer is software: what counts as a “hold” versus a real commitment, how refundable inventory is fenced, and what proof or signals an agent must provide when it books, changes, or cancels.

One hard question to take into your next revenue meeting: if 10% of demand becomes cancellable options held by software, what breaks first in your forecasting, your pricing, or your operations?

Trusted Agents creates production-shaped prototypes

If you’re a travel operator trying to make sense of this, the first step is to stop treating “agentic” as a UI upgrade. Trusted Agents helps leadership teams turn the shift into a clear business case, with a small set of bets you can defend internally, then translate that into a governed pilot with real operational and data requirements.

And we don’t leave you with a deck. We bring a world-class ecosystem of partners to build a production-shaped prototype, with identity, controls, and evidence designed in from day one.

Start here: Trusted Agents​

Want the next Situation Room note?

​The Trusted Agents Ltd.​
​Unsubscribe · Preferences​